关键词: 自然抠图, 透明度估计, 鲁棒抠图, 图拉普拉斯

Abstract: Alpha matting, aiming at extraction of foreground elements from a natural image by means of opacity（alpha）estimation, is one of the key techniques for image editing and film production. However, matting is inherently an ill-posed problem. Many matting approaches perform poorly with the complex natural images. An improved robust matting algorithm is proposed. The proposed method involves labeling some unknown pixels according to the known regions of trimap, selection of foreground and background pairs for unknown pixels to determine of the initial alpha and its confidence, optimization of the initial alpha by minimizing the quadratic object function based on matte Laplacian. The experiments on the benchmark images are carried out and the results show that the proposed method performs better in the qualitative and quantitative comparison with some popular matting algorithms.

Key words: natural matting, alpha estimation, robust matting, matte Laplacian